op_neon.h.svn-base

我们自己开发的一个OSEK操作系统！不知道可不可以？

NEON_POP(pmax_s8, neon_s8, 4)
NEON_POP(pmax_u8, neon_u8, 4)
NEON_POP(pmax_s16, neon_s16, 2)
NEON_POP(pmax_u16, neon_u16, 2)
#undef NEON_FN

NEON_OP(max_f32)
{
    float32 f0 = vfp_itos(T0);
    float32 f1 = vfp_itos(T1);
    T0 = (float32_compare_quiet(f0, f1, NFS) == 1) ? T0 : T1;
    FORCE_RET();
}

#define NEON_FN(dest, src1, src2) dest = (src1 < src2) ? src1 : src2
NEON_VOP(min_s8, neon_s8, 4)
NEON_VOP(min_u8, neon_u8, 4)
NEON_VOP(min_s16, neon_s16, 2)
NEON_VOP(min_u16, neon_u16, 2)
NEON_VOP(min_s32, neon_s32, 1)
NEON_VOP(min_u32, neon_u32, 1)
NEON_POP(pmin_s8, neon_s8, 4)
NEON_POP(pmin_u8, neon_u8, 4)
NEON_POP(pmin_s16, neon_s16, 2)
NEON_POP(pmin_u16, neon_u16, 2)
#undef NEON_FN

NEON_OP(min_f32)
{
    float32 f0 = vfp_itos(T0);
    float32 f1 = vfp_itos(T1);
    T0 = (float32_compare_quiet(f0, f1, NFS) == -1) ? T0 : T1;
    FORCE_RET();
}

#define NEON_FN(dest, src1, src2) \
    dest = (src1 > src2) ? (src1 - src2) : (src2 - src1)
NEON_VOP(abd_s8, neon_s8, 4)
NEON_VOP(abd_u8, neon_u8, 4)
NEON_VOP(abd_s16, neon_s16, 2)
NEON_VOP(abd_u16, neon_u16, 2)
NEON_VOP(abd_s32, neon_s32, 1)
NEON_VOP(abd_u32, neon_u32, 1)
#undef NEON_FN

NEON_OP(abd_f32)
{
    float32 f0 = vfp_itos(T0);
    float32 f1 = vfp_itos(T1);
    T0 = vfp_stoi((float32_compare_quiet(f0, f1, NFS) == 1)
                  ? float32_sub(f0, f1, NFS)
                  : float32_sub(f1, f0, NFS));
    FORCE_RET();
}

#define NEON_FN(dest, src1, src2) dest = src1 + src2
NEON_VOP(add_u8, neon_u8, 4)
NEON_VOP(add_u16, neon_u16, 2)
NEON_POP(padd_u8, neon_u8, 4)
NEON_POP(padd_u16, neon_u16, 2)
#undef NEON_FN

NEON_OP(add_f32)
{
    T0 = vfp_stoi(float32_add(vfp_itos(T0), vfp_itos(T1), NFS));
    FORCE_RET();
}

#define NEON_FN(dest, src1, src2) dest = src1 - src2
NEON_VOP(sub_u8, neon_u8, 4)
NEON_VOP(sub_u16, neon_u16, 2)
#undef NEON_FN

NEON_OP(sub_f32)
{
    T0 = vfp_stoi(float32_sub(vfp_itos(T0), vfp_itos(T1), NFS));
    FORCE_RET();
}

#define NEON_FN(dest, src1, src2) dest = src2 - src1
NEON_VOP(rsb_u8, neon_u8, 4)
NEON_VOP(rsb_u16, neon_u16, 2)
#undef NEON_FN

NEON_OP(rsb_f32)
{
    T0 = vfp_stoi(float32_sub(vfp_itos(T1), vfp_itos(T0), NFS));
    FORCE_RET();
}

#define NEON_FN(dest, src1, src2) dest = src1 * src2
NEON_VOP(mul_u8, neon_u8, 4)
NEON_VOP(mul_u16, neon_u16, 2)
#undef NEON_FN

NEON_OP(mul_f32)
{
    T0 = vfp_stoi(float32_mul(vfp_itos(T0), vfp_itos(T1), NFS));
    FORCE_RET();
}

NEON_OP(mul_p8)
{
    T0 = helper_neon_mul_p8(T0, T1);
}

#define NEON_FN(dest, src1, src2) dest = (src1 & src2) ? -1 : 0
NEON_VOP(tst_u8, neon_u8, 4)
NEON_VOP(tst_u16, neon_u16, 2)
NEON_VOP(tst_u32, neon_u32, 1)
#undef NEON_FN

#define NEON_FN(dest, src1, src2) dest = (src1 == src2) ? -1 : 0
NEON_VOP(ceq_u8, neon_u8, 4)
NEON_VOP(ceq_u16, neon_u16, 2)
NEON_VOP(ceq_u32, neon_u32, 1)
#undef NEON_FN

#define NEON_QDMULH16(dest, src1, src2, round) do { \
    uint32_t tmp = (int32_t)(int16_t) src1 * (int16_t) src2; \
    if ((tmp ^ (tmp << 1)) & SIGNBIT) { \
        env->QF = 1; \
        tmp = (tmp >> 31) ^ ~SIGNBIT; \
    } \
    tmp <<= 1; \
    if (round) { \
        int32_t old = tmp; \
        tmp += 1 << 15; \
        if ((int32_t)tmp < old) { \
            env->QF = 1; \
            tmp = SIGNBIT - 1; \
        } \
    } \
    dest = tmp >> 16; \
    } while(0)
#define NEON_FN(dest, src1, src2) NEON_QDMULH16(dest, src1, src2, 0)
NEON_VOP(qdmulh_s16, neon_s16, 2)
#undef NEON_FN
#define NEON_FN(dest, src1, src2) NEON_QDMULH16(dest, src1, src2, 1)
NEON_VOP(qrdmulh_s16, neon_s16, 2)
#undef NEON_FN
#undef NEON_QDMULH16

#define SIGNBIT64 ((uint64_t)1 << 63)
#define NEON_QDMULH32(dest, src1, src2, round) do { \
    uint64_t tmp = (int64_t)(int32_t) src1 * (int32_t) src2; \
    if ((tmp ^ (tmp << 1)) & SIGNBIT64) { \
        env->QF = 1; \
        tmp = (tmp >> 63) ^ ~SIGNBIT64; \
    } else { \
        tmp <<= 1; \
    } \
    if (round) { \
        int64_t old = tmp; \
        tmp += (int64_t)1 << 31; \
        if ((int64_t)tmp < old) { \
            env->QF = 1; \
            tmp = SIGNBIT64 - 1; \
        } \
    } \
    dest = tmp >> 32; \
    } while(0)
#define NEON_FN(dest, src1, src2) NEON_QDMULH32(dest, src1, src2, 0)
NEON_VOP(qdmulh_s32, neon_s32, 1)
#undef NEON_FN
#define NEON_FN(dest, src1, src2) NEON_QDMULH32(dest, src1, src2, 1)
NEON_VOP(qrdmulh_s32, neon_s32, 1)
#undef NEON_FN
#undef NEON_QDMULH32

NEON_OP(recps_f32)
{
    T0 = vfp_stoi(helper_recps_f32(vfp_itos(T0), vfp_itos(T1)));
    FORCE_RET();
}

NEON_OP(rsqrts_f32)
{
    T0 = vfp_stoi(helper_rsqrts_f32(vfp_itos(T0), vfp_itos(T1)));
    FORCE_RET();
}

/* Floating point comparisons produce an integer result.  */
#define NEON_VOP_FCMP(name, cmp) \
NEON_OP(name) \
{ \
    if (float32_compare_quiet(vfp_itos(T0), vfp_itos(T1), NFS) cmp 0) \
        T0 = -1; \
    else \
        T0 = 0; \
    FORCE_RET(); \
}

NEON_VOP_FCMP(ceq_f32, ==)
NEON_VOP_FCMP(cge_f32, >=)
NEON_VOP_FCMP(cgt_f32, >)

NEON_OP(acge_f32)
{
    float32 f0 = float32_abs(vfp_itos(T0));
    float32 f1 = float32_abs(vfp_itos(T1));
    T0 = (float32_compare_quiet(f0, f1,NFS) >= 0) ? -1 : 0;
    FORCE_RET();
}

NEON_OP(acgt_f32)
{
    float32 f0 = float32_abs(vfp_itos(T0));
    float32 f1 = float32_abs(vfp_itos(T1));
    T0 = (float32_compare_quiet(f0, f1, NFS) > 0) ? -1 : 0;
    FORCE_RET();
}

/* Narrowing instructions.  The named type is the destination type.  */
NEON_OP(narrow_u8)
{
    T0 = (T0 & 0xff) | ((T0 >> 8) & 0xff00)
         | ((T1 << 16) & 0xff0000) | (T1 << 24);
    FORCE_RET();
}

NEON_OP(narrow_sat_u8)
{
    neon_u16 src;
    neon_u8 dest;
#define SAT8(d, s) \
    if (s > 0xff) { \
        d = 0xff; \
        env->QF = 1; \
    } else  { \
        d = s; \
    }

    NEON_UNPACK(neon_u16, src, T0);
    SAT8(dest.v1, src.v1);
    SAT8(dest.v2, src.v2);
    NEON_UNPACK(neon_u16, src, T1);
    SAT8(dest.v3, src.v1);
    SAT8(dest.v4, src.v2);
    NEON_PACK(neon_u8, T0, dest);
    FORCE_RET();
#undef SAT8
}

NEON_OP(narrow_sat_s8)
{
    neon_s16 src;
    neon_s8 dest;
#define SAT8(d, s) \
    if (s != (uint8_t)s) { \
        d = (s >> 15) ^ 0x7f; \
        env->QF = 1; \
    } else  { \
        d = s; \
    }

    NEON_UNPACK(neon_s16, src, T0);
    SAT8(dest.v1, src.v1);
    SAT8(dest.v2, src.v2);
    NEON_UNPACK(neon_s16, src, T1);
    SAT8(dest.v3, src.v1);
    SAT8(dest.v4, src.v2);
    NEON_PACK(neon_s8, T0, dest);
    FORCE_RET();
#undef SAT8
}

NEON_OP(narrow_u16)
{
    T0 = (T0 & 0xffff) | (T1 << 16);
}

NEON_OP(narrow_sat_u16)
{
    if (T0 > 0xffff) {
        T0 = 0xffff;
        env->QF = 1;
    }
    if (T1 > 0xffff) {
        T1 = 0xffff;
        env->QF = 1;
    }
    T0 |= T1 << 16;
    FORCE_RET();
}

NEON_OP(narrow_sat_s16)
{
    if ((int32_t)T0 != (int16_t)T0) {
        T0 = ((int32_t)T0 >> 31) ^ 0x7fff;
        env->QF = 1;
    }
    if ((int32_t)T1 != (int16_t) T1) {
        T1 = ((int32_t)T1 >> 31) ^ 0x7fff;
        env->QF = 1;
    }
    T0 = (uint16_t)T0 | (T1 << 16);
    FORCE_RET();
}

NEON_OP(narrow_sat_u32)
{
    if (T1) {
        T0 = 0xffffffffu;
        env->QF = 1;
    }
    FORCE_RET();
}

NEON_OP(narrow_sat_s32)
{
    int32_t sign = (int32_t)T1 >> 31;

    if ((int32_t)T1 != sign) {
        T0 = sign ^ 0x7fffffff;
        env->QF = 1;
    }
    FORCE_RET();
}

/* Narrowing instructions.  Named type is the narrow type.  */
NEON_OP(narrow_high_u8)
{
    T0 = ((T0 >> 8) & 0xff) | ((T0 >> 16) & 0xff00)
        | ((T1 << 8) & 0xff0000) | (T1 & 0xff000000);
    FORCE_RET();
}

NEON_OP(narrow_high_u16)
{
    T0 = (T0 >> 16) | (T1 & 0xffff0000);
    FORCE_RET();
}

NEON_OP(narrow_high_round_u8)
{
    T0 = (((T0 + 0x80) >> 8) & 0xff) | (((T0 + 0x800000) >> 16) & 0xff00)
        | (((T1 + 0x80) << 8) & 0xff0000) | ((T1 + 0x800000) & 0xff000000);
    FORCE_RET();
}

NEON_OP(narrow_high_round_u16)
{
    T0 = ((T0 + 0x8000) >> 16) | ((T1 + 0x8000) & 0xffff0000);
    FORCE_RET();
}

NEON_OP(narrow_high_round_u32)
{
    if (T0 >= 0x80000000u)
        T0 = T1 + 1;
    else
        T0 = T1;
    FORCE_RET();
}

/* Widening instructions.  Named type is source type.  */
NEON_OP(widen_s8)
{
    uint32_t src;

    src = T0;
    T0 = (uint16_t)(int8_t)src | ((int8_t)(src >> 8) << 16);
    T1 = (uint16_t)(int8_t)(src >> 16) | ((int8_t)(src >> 24) << 16);
}

NEON_OP(widen_u8)
{
    T1 = ((T0 >> 8) & 0xff0000) | ((T0 >> 16) & 0xff);
    T0 = ((T0 << 8) & 0xff0000) | (T0 & 0xff);
}

NEON_OP(widen_s16)
{
    int32_t src;

    src = T0;
    T0 = (int16_t)src;
    T1 = src >> 16;
}

NEON_OP(widen_u16)
{
    T1 = T0 >> 16;
    T0 &= 0xffff;
}

NEON_OP(widen_s32)
{
    T1 = (int32_t)T0 >> 31;
    FORCE_RET();
}

NEON_OP(widen_high_u8)
{
    T1 = (T0 & 0xff000000) | ((T0 >> 8) & 0xff00);
    T0 = ((T0 << 16) & 0xff000000) | ((T0 << 8) & 0xff00);
}

NEON_OP(widen_high_u16)
{
    T1 = T0 & 0xffff0000;
    T0 <<= 16;
}

/* Long operations.  The type is the wide type.  */
NEON_OP(shll_u16)
{
    int shift = PARAM1;
    uint32_t mask;

    mask = 0xffff >> (16 - shift);
    mask |= mask << 16;
    mask = ~mask;

    T0 = (T0 << shift) & mask;
    T1 = (T1 << shift) & mask;
    FORCE_RET();
}

NEON_OP(shll_u64)
{
    int shift = PARAM1;

    T1 <<= shift;
    T1 |= T0 >> (32 - shift);
    T0 <<= shift;
    FORCE_RET();
}

NEON_OP(addl_u16)
{
    uint32_t tmp;
    uint32_t high;

    tmp = env->vfp.scratch[0];
    high = (T0 >> 16) + (tmp >> 16);
    T0 = (uint16_t)(T0 + tmp);
    T0 |= (high << 16);
    tmp = env->vfp.scratch[1];
    high = (T1 >> 16) + (tmp >> 16);
    T1 = (uint16_t)(T1 + tmp);
    T1 |= (high << 16);
    FORCE_RET();
}

NEON_OP(addl_u32)
{
    T0 += env->vfp.scratch[0];
    T1 += env->vfp.scratch[1];
    FORCE_RET();
}

NEON_OP(addl_u64)
{
    uint64_t tmp;
    tmp = T0 | ((uint64_t)T1 << 32);
    tmp += env->vfp.scratch[0];
    tmp += (uint64_t)env->vfp.scratch[1] << 32;
    T0 = tmp;
    T1 = tmp >> 32;
    FORCE_RET();
}

NEON_OP(subl_u16)
{
    uint32_t tmp;
    uint32_t high;

    tmp = env->vfp.scratch[0];
    high = (T0 >> 16) - (tmp >> 16);
    T0 = (uint16_t)(T0 - tmp);
    T0 |= (high << 16);
    tmp = env->vfp.scratch[1];
    high = (T1 >> 16) - (tmp >> 16);
    T1 = (uint16_t)(T1 - tmp);
    T1 |= (high << 16);
    FORCE_RET();
}

NEON_OP(subl_u32)
{
    T0 -= env->vfp.scratch[0];
    T1 -= env->vfp.scratch[1];
    FORCE_RET();
}

NEON_OP(subl_u64)
{
    uint64_t tmp;
    tmp = T0 | ((uint64_t)T1 << 32);
    tmp -= env->vfp.scratch[0];
    tmp -= (uint64_t)env->vfp.scratch[1] << 32;
    T0 = tmp;
    T1 = tmp >> 32;
    FORCE_RET();
}

#define DO_ABD(dest, x, y, type) do { \
    type tmp_x = x; \
    type tmp_y = y; \
    dest = ((tmp_x > tmp_y) ? tmp_x - tmp_y : tmp_y - tmp_x); \
    } while(0)

NEON_OP(abdl_u16)
{
    uint32_t tmp;
    uint32_t low;
    uint32_t high;

    DO_ABD(low, T0, T1, uint8_t);
    DO_ABD(tmp, T0 >> 8, T1 >> 8, uint8_t);
    low |= tmp << 16;
    DO_ABD(high, T0 >> 16, T1 >> 16, uint8_t);
    DO_ABD(tmp, T0 >> 24, T1 >> 24, uint8_t);
    high |= tmp << 16;
    T0 = low;
    T1 = high;
    FORCE_RET();
}

NEON_OP(abdl_s16)
{
    uint32_t tmp;
    uint32_t low;
    uint32_t high;

    DO_ABD(low, T0, T1, int8_t);
    DO_ABD(tmp, T0 >> 8, T1 >> 8, int8_t);
    low |= tmp << 16;
    DO_ABD(high, T0 >> 16, T1 >> 16, int8_t);
    DO_ABD(tmp, T0 >> 24, T1 >> 24, int8_t);
    high |= tmp << 16;
    T0 = low;
    T1 = high;
    FORCE_RET();
}

NEON_OP(abdl_u32)
{
    uint32_t low;
    uint32_t high;

    DO_ABD(low, T0, T1, uint16_t);
    DO_ABD(high, T0 >> 16, T1 >> 16, uint16_t);
    T0 = low;
    T1 = high;
    FORCE_RET();
}

NEON_OP(abdl_s32)
{
    uint32_t low;
    uint32_t high;

    DO_ABD(low, T0, T1, int16_t);
    DO_ABD(high, T0 >> 16, T1 >> 16, int16_t);
    T0 = low;
    T1 = high;
    FORCE_RET();
}

NEON_OP(abdl_u64)
{
    DO_ABD(T0, T0, T1, uint32_t);
    T1 = 0;
}

NEON_OP(abdl_s64)
{
    DO_ABD(T0, T0, T1, int32_t);
    T1 = 0;
}
#undef DO_ABD

/* Widening multiple. Named type is the source type.  */
#define DO_MULL(dest, x, y, type1, type2) do { \
    type1 tmp_x = x; \
    type1 tmp_y = y; \
    dest = (type2)((type2)tmp_x * (type2)tmp_y); \
    } while(0)

NEON_OP(mull_u8)
{
    uint32_t tmp;